Vercel ranks GPT-6.2 Astra first on DeepsecBench
@rauchg says the cybersecurity benchmark uses codebases no lab has had access to for training.
TLDR
Vercel’s leaderboard gives OpenAI the top four DeepsecBench spots, led by GPT-6.2 Astra’s xhigh configuration at 61.22. @rauchg calls it a “stealth” cybersecurity benchmark, saying no lab had access to its codebases for training. He also calls OpenAI’s models the world’s most cybersecurity-capable—a bigger claim than one leaderboard can settle.
Combined views
26.8K
2 Sources, first seen ago

Comments